Investigation of the multifactorial regulatory mechanisms underlying compositional changes in recurrent urinary stones and development of a machine learning-based personalized predictive model.

Investigation of the multifactorial regulatory mechanisms underlying compositional changes in recurrent urinary stones and development of a machine learning-based personalized predictive model - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Search Log in Primary site navigation Search Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Sci Rep . 2026 Apr 10;16:16926. doi: 10.1038/s41598-026-45861-y Investigation of the multifactorial regulatory mechanisms underlying compositional changes in recurrent urinary stones and development of a machine learning-based personalized predictive model Mierzhati Maimaiti 1, # , Waresi Xianmuxiding 2, # , Xinjian Chen 2 , Mahemutijiang Atawula 2 , Guojie Yu 2, ✉ Author information Article notes Copyright and License information 1 Urumqi Friendship Hospital, 558 Shengli Road, Tianshan District, Urumqi, 830002 Xinjiang China 2 Department of Urology, Urumqi Friendship Hospital, 558 Shengli Road, Tianshan District, Urumqi, 830002 Xinjiang China ✉ Corresponding author. # Contributed equally. Received 2025 Nov 8; Accepted 2026 Mar 23; Collection date 2026. © The Author(s) 2026 Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ . PMC Copyright notice PMCID: PMC13230978  PMID: 41963446 Abstract To characterize compositional changes in recurrent urinary calculi, identify associated factors, and develop/validate a prediction model to support risk assessment and individualized prevention. A retrospective cohort compared initial vs. recurrent stone composition and transformation patterns. Patients were classified as composition-unchanged ( n = 113) or composition-changed ( n = 89). Group differences were tested, and predictors were selected using univariate analysis, multivariate logistic regression, and LASSO. Nine machine-learning classifiers were trained and evaluated by AUC/accuracy, with model assessment using calibration, decision curve analysis (DCA), and SHAP. Calcium oxalate stones increased by 12.4% at recurrence, while infectious, uric acid, and other stones decreased by 5.5%, 5.9%, and 0.7% ( P > 0.05). Infectious stones exhibited the highest transformation: only 3.2% remained unchanged, with 45.2% converting to calcium oxalate and 46.8% to uric acid stones. Compared with the changed group, the unchanged group was older and had a higher initial uric acid stone proportion and coronary heart disease prevalence; the changed group had a higher initial calcium oxalate proportion and more hypomagnesemia ( P < 0.05). Multivariate analysis showed that age, initial stone composition, hypocitraturia, hypomagnesuria, and hypomagnesemia were independently associated with composition change, while hyperoxaluria and hyperlipidemia were inversely associated (all P < 0.05). LASSO retained seven core variables. Among the tested models, Random Forest showed the highest performance (training AUC 0.978, accuracy 0.979; test AUC 0.882, accuracy 0.847), with acceptable calibration; DCA suggested a net benefit > 0.6 across thresholds 0.2–0.6. SHAP ranked hyperoxaluria, initial composition, and age as the top contributors. In this cohort, recurrent stone composition change was associated with multiple metabolic and clinical factors, with hypocitraturia, hypomagnesuria, and hypomagnesemia showing independent associations. Among the evaluated algorithms, the Random Forest model achieved the highest performance metrics, and SHAP was used to describe the relative contribution of key variables. Given that this study is a single-center retrospective cohort with a limited sample size, the derived model may exhibit optimistic bias. Its robustness and generalizability therefore require further validation in larger-scale and external multi-center cohorts. Supplementary Information The online version contains supplementary material available at 10.1038/s41598-026-45861-y. Keywords: Recurrent urolithiasis, Stone composition change, Metabolic abnormalities, Machine learning prediction, Random forest Subject terms: Diseases, Medical research, Nephrology, Risk factors, Urology Introduction Urinary calculi are a common condition in urology and represent the most prevalent urological disease in China. Epidemiological studies indicate that the national prevalence of urinary calculi has reached 6.5%, with rates exceeding 10% in certain southern regions 1 . Despite ongoing advancements in minimally invasive surgical techniques 2 , both the incidence and recurrence rates of urinary stones remain high and continue to rise annually 3 – 5 . Evidence suggests that urinary calculi are a chronic, lifelong condition, with reported 5-year and 10-year recurrence rates of up to 50% and 90%, respectively. Effectively reducing disease incidence and preventing recurrence remains a major clinical challenge in urology. The formation mechanisms of urinary stones vary significantly according to their chemical composition, necessitating distinct preventive strategies. Tailoring prevention based on stone composition has been shown to reduce recurrence rates to 10%-15%, making it a cornerstone of effective secondary prevention. However, stone composition in individual patients is not static. Research has demonstrated that major stone types-including calcium oxalate, calcium phosphate, and uric acid stones-can undergo interconversion 6 , 7 . Nevertheless, the patterns and underlying mechanisms of such compositional transitions remain incompletely understood. Although some studies suggest that factors such as stone recurrence interval may influence composition change 8 , the relationship between compositional shifts and metabolic abnormalities, as well as the key determinants driving these transformations, remains poorly defined 9 . In the field of predictive modeling, researchers worldwide have developed various tools for urinary stone prediction, including logistic regression models based on clinical variables and machine learning–enhanced approaches such as support vector machines (SVM) 10 – 12 . These models are primarily designed to predict either the risk of stone recurrence or the composition of initial stones. However, existing models exhibit several notable limitations. First, research focus remains narrow, with most studies centered on binary recurrence prediction (“recurrence vs. non-recurrence”), while few specifically address the dynamic shift in stone composition 13 . Second, model evaluation is often limited to conventional performance metrics-such as AUC, accuracy, sensitivity, and specificity-without incorporating assessments of clinical utility (e.g., DCA) or model interpretability (e.g., SHAP analysis) 14 . Third, many models are derived from small sample sizes, which compromises their generalizability and limits their accuracy and applicability in real-world clinical settings, thereby hindering their use in personalized management strategies 15 . Although prior studies have investigated stone composition shifts, existing predictive models suffer from limited sample sizes and reliance on narrow evaluation metrics. Consequently, research specifically addressing dynamic changes in stone composition remains scarce-this study addresses that gap. Nevertheless, the relatively modest sample size ( n = 202) may constrain the model’s generalizability, underscoring the need for external validation in larger, diverse cohorts. To address these gaps, this study conducted a retrospective analysis of clinical data from 202 patients with recurrent urinary calculi (113 in the unchanged-composition group and 89 in the changed-composition group). We not only investigated the patterns of stone composition changes and their associations with metabolic abnormalities but also innovatively constructed nine classification models. These models were rigorously evaluated across multiple dimensions: predictive performance (AUC, accuracy, sensitivity, specificity), clinical usefulness (decision curve analysis, DCA), calibration (calibration plots), and interpretability (SHAP analysis). The optimal model was selected based on comprehensive assessment, aiming to provide a reliable tool for the precise identification of high-risk individuals for stone composition change and to support the development of individualized intervention strategies. Data selection and processing Data sources Data for this study were collected from patients with recurrent urinary stones who presented to the Friendship Hospital of Urumqi between January 2011 and January 2025. The Friendship Hospital of Urumqi is a nationally recognized tertiary Class A hospital that integrates medical care, disease prevention, teaching, scientific research, health promotion, and rehabilitation services. Its Department of Urology possesses extensive clinical expertise and advanced medical infrastructure in the diagnosis and management of urolithiasis, thereby ensuring access to comprehensive and high-quality patient data. Clinical information was systematically retrieved from the hospital’s electronic medical record system, institutional clinical database, and patient follow-up records. Collected data encompassed demographic characteristics, medical history, laboratory and imaging test results, treatment details, and stone composition analysis reports. Inclusion and exclusion criteria Inclusion criteria: patients with urinary calculi who experienced at least one documented recurrence, with stone composition analysis performed following each treatment episode. This criterion ensures that all included individuals exhibit the defining features of recurrent stone disease and provide analyzable compositional data, thereby enabling a comprehensive investigation into the factors associated with changes in stone composition. Exclusion criteria: patients with incomplete clinical records, those with residual stones present prior to recurrence, and individuals with confirmed hereditary forms of nephrolithiasis were excluded. Incomplete clinical data may result in missing or inaccurate information, compromising data integrity and study reliability. The presence of residual stones before recurrence could confound the accurate assessment of true recurrence and potential compositional shifts. Furthermore, patients with established hereditary stone disorders often have distinct pathophysiological mechanisms underlying stone formation compared to those with non-hereditary disease. To maintain cohort homogeneity and enhance the validity of analytical outcomes, these individuals were excluded from the study. Observation indicators Collection of general information The collection of general patient information was conducted through the hospital’s comprehensive electronic medical record system. Gender was recorded as a binary categorical variable. Age at first diagnosis was extracted from clinical documentation and reported in completed years. Recurrence interval-the time between initial presentation and subsequent recurrence-was calculated using standardized date-difference algorithms to ensure precision. The total number of recurrences was determined by systematically reviewing all documented stone episodes and corresponding diagnostic records, ensuring accurate event counting. Body mass index (BMI) was calculated for each available visit using measured weight (in kilograms) obtained via calibrated digital scales and height (in meters) measured with a fixed stadiometer. BMI was derived using the standard formula: BMI=weight (kg)/height (m)². These baseline demographic and anthropometric variables served as essential covariates for investigating potential determinants of stone composition change. For instance, comparing compositional shifts across gender groups may reveal sex-specific patterns in stone evolution; analysis of age at onset and recurrence interval may help elucidate the influence of developmental stage and temporal dynamics on changes in stone mineralogy.​ Statistical analysis of past medical history and comorbidities The assessment of past medical history and comorbidities was conducted through a comprehensive review of patients’ medical records, including outpatient and inpatient charts, as well as relevant diagnostic reports. The presence of urinary tract obstruction or anatomical malformations was determined based on imaging findings from renal ultrasound and computed tomography urography (CTU). Initial evidence of dilation, stenosis, or structural abnormalities on ultrasound was further evaluated using CTU to precisely localize and characterize the extent of obstruction or congenital anomalies. Diagnosis of diabetes mellitus was established according to standard clinical criteria: fasting plasma glucose ≥ 7.0 mmol/L, 2-hour postprandial glucose ≥ 11.1 mmol/L with classic symptoms (polydipsia, polyphagia, polyuria, and unexplained weight loss), or glycated hemoglobin (HbA1c) ≥ 6.5%, supplemented by documentation of antidiabetic therapy. Hyperparathyroidism was diagnosed when elevated parathyroid hormone (PTH) levels were accompanied by hypercalcemia and hypophosphatemia, with confirmatory imaging-such as parathyroid ultrasound or radionuclide scintigraphy-revealing parathyroid hyperplasia or adenoma. Coronary artery disease was defined based on a combination of clinical symptoms, electrocardiographic evidence of ischemia (ST-T wave changes), and/or echocardiographic demonstration of regional wall motion abnormalities. Hypertension was identified if systolic blood pressure was ≥ 140 mmHg and/or diastolic blood pressure ≥ 90 mmHg across multiple measurements, or if the patient had a documented history of hypertension and was undergoing antihypertensive treatment. Systematic documentation of these comorbid conditions enabled rigorous evaluation of their potential associations with changes in stone composition, thereby supporting more informed clinical decision-making in the management and prevention of recurrent urolithiasis. Detection of infection and metabolic indicators​ The diagnosis of urinary tract infection (UTI) was established based on standardized criteria incorporating urinalysis and urine bacterial culture results. Microscopic urinalysis showing pyuria-defined as > 5 white blood cells per high-power field-or the presence of pus cells, in conjunction with clinical symptoms such as dysuria, frequency, urgency, or flank pain, prompted further evaluation via urine culture. Clean-catch midstream urine samples were collected under sterile conditions, inoculated onto appropriate culture media, and incubated under controlled temperature and atmospheric conditions. A colony count exceeding 10⁵ colony-forming units per milliliter (CFU/mL) was considered diagnostic of UTI; for symptomatic patients, counts ≥ 10² CFU/mL were also deemed clinically significant. In cases involving cocci, a threshold of 10³-10⁴CFU/mL was considered indicative of infection. Additional supportive evidence included a one-hour urinary leukocyte excretion rate > 200,000 cells, peripheral leukocytosis, neutrophilic left shift, and elevated erythrocyte sedimentation rate (ESR), which were used as adjunctive indicators to support the diagnosis. For 24-hour urine metabolic profiling, patients were instructed to collect all urine over a 24-hour period, with total volume recorded. Urinary electrolytes-including sodium, potassium, calcium, and magnesium-were measured using the ion-selective electrode (ISE) method, which determines ion concentration based on the electrode’s selective electrochemical response. Quantification of urinary oxalate, uric acid, and citrate was performed by high-performance liquid chromatography (HPLC), a technique capable of resolving complex mixtures into individual components and enabling precise quantification through comparison with calibrated reference standards.Blood samples were drawn after an overnight fast to assess metabolic parameters. Parathyroid hormone (PTH) levels were determined using chemiluminescent immunoassay (CLIA), a highly sensitive method that employs labeled antibodies and measures emitted light intensity to quantify analyte concentration. Serum electrolytes were analyzed using an automated biochemical analyzer, employing colorimetric and ISE methodologies. Serum uric acid was measured enzymatically via uricase-catalyzed oxidation, with subsequent colorimetric detection of reaction products. Lipid profiles-including total cholesterol (TC), triglycerides (TG), low-density lipoprotein cholesterol (LDL-C), and high-density lipoprotein cholesterol (HDL-C)-were also assessed using the same platform.Metabolic abnormalities in both serum and 24-hour urine were defined according to reference ranges recommended by the European Association of Urology (EAU) guidelines on urolithiasis. Hyperlipidemia was diagnosed if any of the following criteria were met: TC ≥ 5.2 mmol/L, TG ≥ 1.7 mmol/L, LDL-C ≥ 3.4 mmol/L, or HDL-C ≤ 1.0 mmol/L in fasting venous blood. Analysis of stone composition​ The analysis of stone composition is conducted using infrared spectroscopy. The principle is based on the characteristic absorption peaks of the stone samples in the infrared light region to determine the structure and composition of the compounds. Different types of stones have different absorption wavelengths of infrared rays. By detecting the absorption of the stones by infrared rays and comparing it with the standard spectra of known stone compositions, the composition of human stones can be analyzed. The specific operation process is as follows: After the patient undergoes stone treatment, collect the stones excreted by the patient, the stones after fragmentation and recrystallization, or the stones obtained from surgery during the operation. Clean the surface of the stone specimens with clean water, then place them in a drying oven and dry them at 50 °C for 3–4 h, or leave them at room temperature to dry naturally and store them for later use. Take 1 mg of the dried stone sample powder, mix it thoroughly with 200 mg of pure potassium bromide that has been fully dried in a agate mortar, and grind it evenly. Use a press machine to compress the mixed sample to 20 MPa and keep it for 3 min to make a semi-transparent sheet. Quickly place the made semi-transparent sheet into the sample chamber of the infrared spectrometer for scanning, obtaining the infrared spectrum of the stone. Through computer retrieval of the spectral library established by using pure substances in different ratios to simulate various urinary stone components, conduct qualitative analysis based on the characteristics of the sample infrared spectrum to determine the main components contained in the stone. Then, use the absorbance ratio method for quantitative analysis to determine the specific proportion of each component in the stone. Based on the main components of the stones (the proportion of which is ≥ 50%), the stones are classified into calcium oxalate stones (monohydrate and dihydrate calcium oxalate), infectious stones (calcium phosphate carbonate, magnesium ammonium phosphate hexahydrate, hydroxyapatite), uric acid stones (anhydrous uric acid, monohydrate and dihydrate sodium urate, urate ammonium), and other types of stones (phosphate calcium stones, calcite, cystine stones). If the main components of the stones change, it is defined as a change in stone composition. For example, if the proportion of calcium oxalate in the initial stone composition is ≥ 50%, but the proportion is < 50% in the recurrence, and the proportions of other components have significantly changed, it is determined as a change in stone composition. This accurate method of stone composition analysis provides precise classification basis for subsequent research on the influencing factors of stone composition changes, and helps to formulate targeted prevention and treatment strategies. Statistical analysis and model construction Statistical analysis was conducted using SPSS 26.0 and R software. After normality testing, measurement data were expressed as mean±standard deviation ( ± s), and intergroup comparisons were performed using independent sample t-tests. Count data were expressed as number (percentage), and intergroup comparisons were conducted using χ² tests or Fisher’s exact tests. First, univariate analysis was performed to screen candidate variables related to changes in stone composition, followed by multivariate logistic regression to identify independent influencing factors. To reduce collinearity and optimize feature dimensions, LASSO regression was further applied within the training set for feature selection, and nine classification models (Logistic regression, SVM, GBM, neural network, random forest, XGBoost, KNN, AdaBoost, CatBoost) were constructed based on the selected variables. Model performance was evaluated using metrics such as AUC, accuracy, sensitivity, and specificity, and calibration curves and decision curve analysis (DCA) were used to assess calibration and clinical net benefit. SHAP was used to enhance model interpretability. To avoid data leakage, the data were stratified randomly divided into training and test sets at a ratio of 7:3. All preprocessing and feature selection were only performed on the training set, and the parameters fitted from the training set were applied to the test set for independent evaluation. Considering the risk of overestimating performance due to limited sample size, internal validation by bootstrap (1000 times) was further conducted on the final model to correct for optimistic bias. To improve the readability of the main text, details regarding model hyperparameter settings, the LASSO cross-validation process, missing data handling rules, and the implementation details of bootstrap internal validation (including code and parameter tables) have been moved to the supplementary materials (Methods S1–S3). Results Composition ratio of initial and recurrent stones An analysis of the compositional distribution between initial and recurrent urinary stones (Fig. 1 ) revealed that, compared with initial episodes, the proportion of calcium oxalate stones increased by 12.4% upon recurrence, whereas the proportions of infectious stones, uric acid stones, and other stone types decreased by 5.5%, 5.9%, and 0.7%, respectively. However, the overall differences in stone composition between the two groups were not statistically significant ( P > 0.05). Fig. 1. Open in a new tab Composition ratio of initial and recurrent stone components in 202 patients with recurrent urinary calculi. Changes in stone composition during recurrence Calcium oxalate stones accounted for 75.8% of the total, while uric acid stones accounted for 52.4%. The non-self components of both were relatively low, with the components mainly being the self-components and the degree of change being small. Infectious stones accounted for only 3.2% of the total, and over 95% of the components were other types such as calcium oxalate and uric acid stones. The deviation from the self-components was severe. The components of other types of stones could be ignored, and they were almost entirely composed of calcium oxalate and uric acid stones, with the component changes reaching the extreme. In summary, in this study cohort, the component changes of other types of stones were the most prominent, and the degree of change of infectious stones was also significant. The components of calcium oxalate and uric acid stones were mainly self-components, with less change.(Fig. 2 ) . Fig. 2. Open in a new tab The interconversion situation among different stone components in 202 patients with recurrent urinary calculi. Analysis of general data This study compared the clinical characteristics between the group with unchanged stone composition ( n = 113) and the group with changed stone composition ( n = 89). The results showed that in the indicators with significant statistical differences ( P < 0.05), the average age of the group with unchanged stone composition (42.9 ± 10.1 years), the initial proportion of uric acid stones (30.1%), the recurrence proportion of uric acid stones (18.6%), the incidence of coronary heart disease (9.7%), the prevalence of hyperoxaluria (51.3%), and the prevalence of hyperlipidemia (43.4%) were significantly higher in the group with unchanged stone composition than in the group with changed stone composition; while the initial proportion of calcium oxalate stones in the group with changed stone composition (56.2%), the prevalence of hypomagnesuria (13.5%), hypomagnesemia (16.9%), and hypouricemicuria (15.7%) were significantly higher. There were no significant differences in multiple indicators such as gender, frequency of attacks, and urinary tract obstruction between the two groups ( P > 0.05). In conclusion, the results of this study suggest that age, initial stone type, magnesium metabolism, and citrate levels may be key factors influencing the change in stone composition. Younger individuals, those with initial calcium oxalate stones and combined with magnesium metabolism abnormalities, and those with low citrate urine are relatively more prone to have a change in stone composition (Table 1 ). Table 1. Comparison of baseline data between the two groups of patients. Indicator Category Unchanged stone composition group ( N = 113) Changed stone composition group ( N = 89) P Value Gender (n, % ) Male 88 (77.9%) 76 (85.4%) 0.24 Female 25 (22.1%) 13 (14.6%) Age (year, mean ± SD ) 42.9 ± 10.1 38.7 ± 9.6 0.003 Number of episodes (n, % ) One 109 (96.5%) 80 (89.9%) 0.109 Two 4 (3.5%) 9 (10.1%) Initial onset (n, % ) Calcium oxalate stones 41 (36.3%) 50 (56.2%) < 0.001 Infectious stones 38 (33.6%) 24 (27.0%) Uric acid stones 34 (30.1%) 8 (9.0%) Other types of stones 0 (0%) 7 (7.9%) Recurrence (n, % ) Calcium oxalate stones 64 (56.6%) 52 (58.4%) 0.032 Infectious stones 28 (24.8%) 23 (25.8%) Uric acid stones 21 (18.6%) 9 (10.1%) Other types of stones 0 (0%) 5 (5.6%) Urinary obstruction (n, % ) No 106 (93.8%) 82 (92.1%) 0.853 Yes 7 (6.2%) 7 (7.9%) Hypertension (n, % ) No 97 (85.8%) 80 (89.9%) 0.514 Yes 16 (14.2%) 9 (10.1%) Diabetes mellitus (n, % ) No 96 (85.0%) 82 (92.1%) 0.178 Yes 17 (15.0%) 7 (7.9%) Coronary heart disease (n, % ) No 102 (90.3%) 88 (98.9%) 0.023 Yes 11 (9.7%) 1 (1.1%) BMI (n, % ) < 23.9 16 (14.2%) 17 (19.1%) 0.061 24—27.9 78 (69.0%) 47 (52.8%) > 28 19 (16.8%) 25 (28.1%) Hyperparathyroidism (n, % ) No 109 (96.5%) 89 (100%) 0.199 Yes 4 (3.5%) 0 (0%) Urinary tract infection (n, % ) No 70 (61.9%) 58 (65.2%) 0.745 Yes 43 (38.1%) 31 (34.8%) Urine specific gravity (n, % ) >2500 2 (1.8%) 0 (0%) 0.385 1500–2500 19 (16.8%) 18 (20.2%) <1500 92 (81.4%) 71 (79.8%) Hyperoxaluria (n, % ) No 55 (48.7%) 72 (80.9%) < 0.001 Yes 58 (51.3%) 17 (19.1%) Hypernatriuria (n, % ) No 99 (87.6%) 77 (86.5%) 0.985 Yes 14 (12.4%) 12 (13.5%) Hypercalciuria (n, % ) No 100 (88.5%) 75 (84.3%) 0.504 Yes 13 (11.5%) 14 (15.7%) Hyperphosphaturia (n, % ) No 109 (96.5%) 80 (89.9%) 0.109 Yes 4 (3.5%) 9 (10.1%) Hypomagnesuria (n, % ) No 111 (98.2%) 77 (86.5%) 0.003 Yes 2 (1.8%) 12 (13.5%) Hyperuricosuria (n, % ) No 108 (95.6%) 82 (92.1%) 0.467 Yes 5 (4.4%) 7 (7.9%) Hypocitraturia (n, % ) No 109 (96.5%) 75 (84.3%) 0.006 Yes 4 (3.5%) 14 (15.7%) Hypercalcemia (n, % ) No 102 (90.3%) 79 (88.8%) 0.908 Yes 11 (9.7%) 10 (11.2%) Hypomagnesemia (n, % ) No 109 (96.5%) 74 (83.1%) 0.003 Yes 4 (3.5%) 15 (16.9%) Hyperphosphatemia (n, % ) No 111 (98.2%) 85 (95.5%) 0.475 Yes 2 (1.8%) 4 (4.5%) Hyperuricemia (n, % ) No 106 (93.8%) 80 (89.9%) 0.447 Yes 7 (6.2%) 9 (10.1%) Hyperlipidemia (n, % ) No 64 (56.6%) 71 (79.8%) < 0.001 Yes 49 (43.4%) 18 (20.2%) Metabolic syndrome (n, % ) No 101 (93.5%) 67 (85.9%) 0.138 Yes 7 (6.5%) 11 (14.1%) Open in a new tab Univariate and multivariate logistic regression analysis In the single-factor analysis, we tested multiple potentially related indicators, and the results showed that there were 7 indicators that were significantly associated with the change in stone composition ( P < 0.05). The OR value of age was 0.96 (95%CI: 0.93–0.99, P = 0.004), indicating that for each additional year of age in the statistical model of this study, the risk of stone composition change statistically decreased by 4%. This suggests that age growth may have a certain inhibitory effect on the change of stone composition. The OR value of stone composition was 0.71 (95%CI: 0.51–0.99, P = 0.043), indicating that the initial stone composition might be related to the risk of composition change. The OR values of coronary heart disease, hyperoxaluria, and hyperlipidemia were all less than 1, and they were protective factors for stone composition change in this study; the OR values of hypomagnesuria, hypouricemia, and hypomagnesemia were greater than 1, and they were risk factors in this study. The results of the multivariate Logistic regression analysis showed that after screening out the 7 indicators that might affect the change in stone composition through the single-factor analysis, we further clarified the independent effects of these factors by including them in the multivariate Logistic regression analysis. The analysis results indicated that among these 7 indicators, coronary heart disease had no statistical significance ( P = 0.327), while the other 6 indicators became independent influencing factors for the change in stone composition in the model set by this study ( P < 0.05). Age was an independent protective factor, with an OR value of 0.94 (95% CI: 0.90–0.98, P = 0.005), indicating that in this study cohort, the older the age, the lower the risk of stone composition change. The OR value of stone composition was 0.48 (95% CI: 0.30–0.76, P = 0.002), confirming its independent influence on the change in stone composition. The OR values of hyperoxaluria and hyperlipidemia were both less than 1, and they were independent protective factors in this study; while hypomagnesuria, hypouricemia, and hypomagnesemia had OR values greater than 1, and they were independent risk factors in this study, with the most significant increase in the risk of stone composition change (Table 2 ). Table 2. Univariate and multivariate logistic regression analysis of factors affecting changes in stone composition. Variable Univariate analysis Multivariate analysis OR value 95% confidence interval (95% CI) P value OR value 95% confidence interval (95% CI) P value Age 0.96 0.93–0.99 0.004 0.94 0.90–0.98 0.005 Stone composition 0.71 0.51–0.99 0.043 0.48 0.30–0.76 0.002 Coronary heart disease 0.11 0.01–0.83 0.033 0.32 0.03–3.11 0.327 Hyperoxaluria 0.22 0.12–0.43 < 0.001 0.1 0.04–0.26 < 0.001 Hypomagnesuria 8.65 1.88–39.74 0.006 12.42 2.12–72.64 0.005 Hypocitraturia 5.09 1.61–16.06 0.005 56.42 9.30-342.50 < 0.001 Hypomagnesemia 5.52 1.76–17.30 0.003 10.16 1.91–54.11 0.007 Hyperlipidemia 0.33 0.18–0.63 < 0.001 0.28 0.12–0.65 0.003 Open in a new tab Use LASSO regression to select the variables to be included in the model​ LASSO regression achieves this by introducing L1 regularization (i.e., the λ penalty coefficient), which can compress the weights of less significant variables in the coefficient vector to 0. This process enables the identification of the features that have the greatest predictive power for the target variable, thereby simplifying the model and enhancing its generalization ability. In this study, all the variables selected through the aforementioned screening process were included in the LASSO regression analysis. The core variables corresponding to the closest distance of the mean of the binomial deviation (1 standard error) were chosen for subsequent model construction. The results of the LASSO regression analysis showed that Age, Stone Composition, Hyperoxaluria, Hypomagnesuria, Hypocitraturia, Hypomagnesemia, and Hyperlipidemia were independent risk factors for the occurrence of changed recurrence stone components under the screening criteria of this study (Fig. 3 ). Fig. 3. Open in a new tab LASSO Regression a: LASSO Regression Cross-validation Curve; b: LASSO Coefficient Path Plot. Performance of different classification models on training set and test set To accurately predict the risk of recurrence of stone composition changes, this study systematically constructed and evaluated nine different machine learning classification models, including Logistic Regression, SVM, GBM, Neural Network, Random Forest, Xgboost, KNN, Adaboost, and CatBoost. We comprehensively evaluated the model performance using six indicators such as AUC (area under the curve), accuracy, sensitivity, and specificity (Table 3 ). Table 3. Comparison of performance indicators of different machine learning classification models on the training set and test set. Data set Classification model AUC Accuracy Sensitivity Specificity Precision F1 score Training set Logistic 0.869 0.797 0.714 0.863 0.804 0.756 SVM 0.855 0.776 0.841 0.725 0.707 0.768 GBM 0.744 0.72 0.619 0.8 0.709 0.661 NeuralNetwork 0.860 0.783 0.778 0.787 0.742 0.76 RandomForest 0.978 0.979 0.968 0.988 0.984 0.976 Xgboost 0.828 0.755 0.746 0.762 0.712 0.729 KNN 0.972 0.972 0.968 0.975 0.968 0.968 Adaboost 0.784 0.727 0.825 0.65 0.65 0.727 CatBoost 0.923 0.888 0.825 0.938 0.912 0.867 Test set Logistic 0.838 0.814 0.846 0.788 0.759 0.8 SVM 0.797 0.797 0.846 0.758 0.733 0.786 GBM 0.758 0.712 0.923 0.545 0.615 0.738 NeuralNetwork 0.838 0.814 0.808 0.818 0.778 0.792 RandomForest 0.882 0.847 0.846 0.848 0.815 0.83 Xgboost 0.807 0.78 0.769 0.788 0.741 0.755 KNN 0.805 0.814 0.731 0.879 0.826 0.776 Adaboost 0.816 0.78 0.885 0.697 0.697 0.78 CatBoost 0.884 0.864 0.885 0.848 0.821 0.852 Open in a new tab In the training set, the Random Forest model demonstrated the most optimal overall predictive performance, with an AUC of 0.978, an accuracy rate of 0.979, and all evaluation indicators exceeding 0.96, showing strong fitting ability. The KNN model performed secondarily, with an AUC of 0.972 and an accuracy rate of 0.972, and all indicators remained above 0.968. The CatBoost model also performed well (AUC 0.923, accuracy 0.888), and its performance was superior to Logistic Regression (AUC 0.869) and Neural Network (AUC 0.860) in the evaluation system of this study. In contrast, the GBM model performed relatively the least well, with an AUC of 0.744 and an accuracy rate of 0.72. In the validation of the test set, the CatBoost model demonstrated relatively the best stability and comprehensive performance, with an AUC of 0.884, an accuracy rate of 0.864, sensitivity of 0.885, and specificity of 0.848. Although the performance of the Random Forest model slightly declined in the test set, it still maintained an excellent level (AUC 0.882, accuracy 0.847), and all indicators were balanced. Logistic Regression and Neural Network performed similarly (both AUC 0.838, both accuracy 0.814), but the former had a higher sensitivity and the latter had a better specificity. The GBM model performed the worst in the test set, with an accuracy rate of only 0.712 and a specificity as low as 0.545, showing a significant imbalance in performance (Fig. 4 ). Fig. 4. Open in a new tab ROC curve analysis of the three models in the training set ( a ) and the validation set ( b ). Based on the performance of the models in the training set and the test set, the Random Forest model was determined as the relatively optimal model for predicting stone composition changes under the conditions set in this study. The CatBoost model became a potential reliable alternative model due to its outstanding stability in the test set. However, the GBM model had relatively lower applicability in this study. Clinical value evaluation of the model based on DCA This study takes the standardized net benefit as the core indicator, combined with the high-risk threshold of 0.0–1.0, and evaluates the clinical value of 9 models and the “full intervention” and “no intervention” strategies through DCA. The results are as follows: In the training set, the Random Forest and KNN models performed relatively best, and the net benefit curve in the full threshold range ranked first. The former had a net benefit of over 0.6 in the 0.2–0.6 threshold range. CatBoost and Logistic Regression performed well, with a net benefit of 0.4–0.5 in the 0.3–0.7 threshold range. GBM and Adaboost were relatively poor, with the net benefit of some thresholds being lower than “no intervention”. The overall trend in the test set was consistent. CatBoost was the most stable and performed the best, with the curve ranking first in the 0.1–0.8 threshold range, and the net benefit fluctuated little in the 0.3–0.6 clinical commonly used threshold range. Random Forest and KNN still had a relatively high net benefit, with a net benefit of 0.4–0.5 in the 0.2–0.7 threshold range. Logistic Regression and Neural Network had good stability. GBM was always lower than “no intervention”, and Adaboost had a narrow application range. In summary, in the evaluation scenario of this study, the net benefit of Random Forest and KNN was the highest in the training set, CatBoost was the most stable in the test set, and all three demonstrated certain clinical value; GBM is not recommended for the corresponding actual decision-making scenario in this study (Fig. 5 ) . Fig. 5. Open in a new tab Analysis of standardized net benefit decision curves for different machine learning models in the training set ( a ) and the test set ( b ). Evaluation of the consistency between model predicted probabilities based on calibration graph and actual observations results This study evaluated the consistency between the predicted probabilities of 10 models (including the newly added LightGBM) and the actual observed results by using calibration graphs (with the “Bin Midpoint” as the horizontal axis and the “Observed Event Percentage” as the vertical axis). The results are as follows: In the training set (TrainCalibration.pdf), the calibration performance of Random Forest, CatBoost, and KNN models was relatively optimal, and the observed event percentages changed linearly with the group midpoints (30–70), closely matching the ideal calibration line; Logistic regression, Neural Network, and LightGBM performed well, with overall small deviations; SVM, Xgboost, and Adaboost had moderate calibration deviations, and there were discrepancies between the observed and predicted values at some group midpoints; GBM was the worst, deviating significantly from the ideal line. In the test set (TestCalibration.pdf), CatBoost still maintained relatively optimal calibration performance, with high matching between observed and predicted values across the entire group range; Random Forest and KNN were second, with slightly more deviations compared to the training set but overall stability; Logistic regression, LightGBM had good calibration consistency; SVM, Xgboost, and Adaboost had increased deviations; GBM still performed the worst, with poor calibration results. In summary, within the data set of this study, CatBoost, Random Forest, and KNN had superior calibration performance in both datasets and relatively higher prediction reliability; GBM had poor calibration results and relatively lower prediction credibility (Fig. 6 ) . Fig. 6. Open in a new tab Calibration graph analysis of different machine learning models in the training set ( a ) and the test set ( b ). Analysis of the importance and mechanism of risk factors in the random forest model based on SHAP analysis In the recurrence prediction of stone composition changes based on the Random Forest model, the importance of each risk factor was initially clarified through SHAP analysis. From the SHAP feature importance graph, it can be seen that the mean (|SHAP value|) of Hyperoxaluria (high oxalate urine) is the highest, and it is a relatively important predictive factor in this model. Next is SC and Age. The importance of Hyperlipidemia is also significant, while Hypomagnesuria, Hypocitraturia, Hypomagnesemia, and CHD have decreasing effects. The SHAP scatter plot further reveals the association between feature values and SHAP values in this model: The high feature value (yellow points) of Hyperoxaluria corresponds to a higher SHAP value, and its positive impact on the model prediction is more significant; The distribution of feature values of SC and Age is significantly correlated with SHAP values, and higher values enhance the contribution to the prediction; The high feature value of Hyperlipidemia also shows a strong positive impact; While Hypomagnesuria, Hypocitraturia, etc., when the feature value is high, the SHAP value is mostly positive, suggesting that these factors may have a promoting effect on the prediction of recurrence of stone composition changes in the high-value state. In summary, in the Random Forest model constructed in this study, high oxalate urine is the core risk factor for predicting recurrence of stone composition changes, and combining factors such as SC and Age may effectively improve the predictive efficacy and interpretability of the model (Fig. 7 ) . Fig. 7. Open in a new tab SHAP analysis of recurrence risk factors for stone composition changes based on the Random Forest model [( a ) feature importance graph; ( b ) scatter correlation graph]. Supplementary note on the completeness of metabolic assessment and potential genetic factors The metabolic-related variables in this study were obtained from the 24-hour urine/serum metabolic abnormality indicators available in the medical records (Methods S1 and Table S1). The database of this study did not have structured records of the completion of the full metabolic assessment, which included “two consecutive 24-hour urine collections + complete serum metabolic profile”, thus the proportion of patients who completed the full metabolic assessment (n_full) could not be calculated. Additionally, although patients with clinically confirmed hereditary stones were excluded, this cohort did not undergo systematic genetic screening. Model performance after bootstrap correction Given the limited sample size and the difference in performance between training and testing suggesting potential overfitting, we conducted bootstrap internal validation (B = 1000) on the final selected Random Forest model and calculated the optimism-corrected discrimination. The apparent AUC of the model was 0.954, with an average optimism bias of 0.040, and the optimism-corrected AUC after correction was 0.914.The ROC curve (Fig. 8 a) illustrates the apparent performance of the final Random Forest model, while Fig. 8b presents the bootstrap distributions of AUC and optimism bias across the iterations. Fig. 8. Open in a new tab Model performance after bootstrap correction [( a ) ROC curve (final random forest) with optimism-corrected AUC; ( b ) bootstrap distribution of AUC and optimism bias]. Discussion The management of urinary calculi has witnessed remarkable advancements in recent years, particularly in minimally invasive surgical techniques, which have proven highly effective in stone removal 16 . Nevertheless, a stark contrast exists in the realm of stone prevention, where recurrence rates remain unacceptably high-reaching up to 50% within a decade. This underscores the inherent limitations of the current treatment paradigm, which predominantly emphasizes surgical clearance without adequately addressing long-term preventive strategies 17 . The pathogenesis of urinary stones is a multifactorial process influenced by a constellation of factors, including race, age, genetic predisposition, environmental exposures, and metabolic status. Among these, metabolic abnormalities are widely acknowledged as central drivers of stone formation 18 , 19 . A substantial body of evidence indicates that the majority of patients with nephrolithiasis exhibit one or more metabolic derangements, such as hyperoxaluria and hyperuricosuria, which significantly elevate the risk of recurrence 20 , 21 . While contemporary interventions effectively eliminate existing stones, they fail to correct the underlying metabolic imbalances that perpetuate disease recurrence. Consequently, the clinical approach to urinary stone disease is undergoing a critical paradigm shift-from a reactive model centered on end-stage stone extraction toward a proactive, comprehensive management strategy that prioritizes early risk stratification, precise metabolic diagnosis, and individualized preventive interventions 22 , 23 . As emphasized by Xu CB et al., achieving an equitable emphasis on “surgery and prevention” is essential, with etiological analysis serving as the cornerstone of precision prevention. In this context, the present study focuses on key indicators that directly reflect alterations in the internal metabolic environment associated with recurrent stone formation—specifically, the dynamic evolution of stone composition. A systematic analysis of influencing factors was conducted, and a precise predictive model was developed to provide potential tool support for the practical implementation of this emerging paradigm. The universality and clinical significance of stone composition changes In this single-center study of 202 patients, stone recurrence was not merely a repetition of the original composition, and changes in stone composition appear to be a common phenomenon 24 . This finding challenges the outdated paradigm of “one-time lithogenesis, lifelong same stone” within the context of this study, suggesting that stone formation may be dynamically regulated by the body’s internal metabolic environment 25 , 26 . Our analysis further identifies that interconversion among calcium oxalate, uric acid, and infection-related stones constitutes the predominant pathway of compositional change. This transformation pattern likely reflects dynamic shifts in the urinary microenvironment driven by factors such as metabolic status, dietary habits, medication use, and urinary tract infections 27 , 28 : a shift from calcium oxalate to uric acid stones may indicate insulin resistance or diet-associated persistent acidic urine, whereas transition to infection-related stones strongly suggests urease-producing bacterial colonization. Therefore, within the clinical framework of this study, changes in stone composition may serve as a valuable biological marker for monitoring alterations in the patient’s internal milieu, offering real-time insights into metabolic dynamics and urinary conditions. These findings provide a direct basis for clinical assessment and personalized adjustment of preventive and therapeutic strategies, demonstrating potential clinical utility. Core driving and protective factors for changes in stone composition Through univariate and multivariate logistic regression analyses, we precisely identified independent factors associated with changes in stone composition within the statistical framework of this study. Age, initial stone composition, hypocitraturia, hypomagnesuria, and hypomagnesemia were determined to be significant independent risk factors. Notably, hypocitraturia exhibited a particularly pronounced effect: as a key inhibitor of crystallization in urine, citrate deficiency directly impairs the body’s capacity to inhibit calcium oxalate crystallization, thereby fostering an unstable urinary chemical environment that may predispose individuals to compositional shifts even under minor metabolic perturbations 29 , 30 . Abnormalities in magnesium metabolism further compromise a secondary inhibitory pathway, acting synergistically to enhance lithogenic instability 31 . Intriguingly, hyperoxaluria and hyperlipidemia demonstrated a paradoxical “protective” association in multivariate analysis-being linked to a lower likelihood of stone composition change. This finding does not diminish their established role in stone formation 32 , 33 , but rather suggests that their pathophysiological mechanisms may promote a sustained and stable lithogenic drive. Specifically, chronic high oxalate excretion may consolidate a calcium oxalate–dominant mineralization pathway, while hyperlipidemia-induced persistent acidic urine may stabilize conditions favorable for uric acid or calcium oxalate stone development, thereby reducing the propensity for phenotypic transition 34 , 35 . These observations imply that certain metabolic disturbances, by establishing a consistent pathogenic milieu, may paradoxically reduce the variability of stone composition. The application value of random forest model and SHAP analysis The various influencing factors exhibit complex interrelationships, rendering traditional statistical models insufficient for accurate individualized risk prediction. In this study, nine machine learning algorithms were systematically introduced and compared. The Random Forest model demonstrated superior and consistent performance in both the training and test sets, with AUC values of 0.978 and 0.882, respectively. Its calibration curve closely aligned with the ideal diagonal, indicating excellent agreement between predicted probabilities and observed outcomes. DCA further confirmed that the model provides clinically meaningful net benefit across a broad range of decision thresholds, suggesting its potential as a reliable tool for identifying patients at high risk of stone composition change 36 . Of particular value, SHAP analysis was employed to elucidate the underlying decision logic of the model, thereby enhancing the interpretability of its predictions 37 . Quantitative assessment of SHAP values revealed hyperoxaluria as the most influential predictor-a finding fully consistent with the established central role of elevated oxalate levels in calcium oxalate stone formation 38 . Age and initial stone type ranked next in importance, reflecting the patient’s dynamic metabolic background and baseline pathological condition, respectively. These results not only corroborate the prior statistical findings but also integrate disparate risk factors into a structured, interpretable, and visually accessible framework, offering actionable insights for the development of personalized and dynamic prevention strategies in clinical practice 39 . Clinical insights and future directions This study has meaningful implications for advancing the paradigm of kidney stone prevention. Young patients experiencing their first episode of calcium oxalate stones and presenting with hypocitraturia or magnesium metabolism abnormalities constitute an “extremely high-risk” population for stone composition change. Clinically, the traditional “one-time analysis, lifelong prevention” approach should be abandoned in favor of incorporating recurrent stone composition analysis 40 and periodic metabolic evaluations 41 into standard clinical practice-a strategy highly consistent with evidence indicating that metabolic assessment can identify the underlying cause in up to 95% of stone formers. Given the intrinsic heterogeneity of urinary stones, it is essential to collect complete calculi or sufficient representative fragments to ensure analytical accuracy and minimize sampling error. The random forest model developed in this study represents a significant step toward transitioning stone prevention and management from a population-based to a personalized approach. Future integration of this model into clinical decision-support systems may enable automated, individualized risk stratification, enhancing preventive precision and clinical efficiency 42 , 43 .The application of machine learning methods to relatively small datasets carries an inherent risk of overfitting, potentially resulting in inflated performance estimates. To mitigate this concern, we employed rigorous internal validation-including 1000-replicate bootstrap resampling-and applied optimism correction to the apparent performance metrics. Nevertheless, these internal assessments cannot substitute for external validation, and the model’s generalizability and robustness warrant confirmation in larger, independent populations. While the model demonstrated strong discrimination and calibration in both the derivation and internal test sets, its clinical utility across diverse real-world settings remains to be established. We therefore plan to prospectively validate the model in multicenter, geographically and temporally distinct cohorts to rigorously assess its transportability and clinical applicability.Although we have adopted bootstrap optimistic bias correction (repeated 1000 times) to enhance the internal validation strength and partially correct the overestimation of performance, this method still belongs to internal resampling validation and cannot replace time-split validation or external independent cohort validation; thus, the transferability and real-world predictive performance of this model still need to be further confirmed in independent cohorts from multiple centers and different time periods.In clinical practice, the model in this study can be used as an auxiliary tool for risk stratification in post-recurrence outpatient settings: for patients predicted to be at high risk of “stone composition change”, it is recommended to prioritize the re-examination of stone composition in the early stage of recurrence and complete the systematic metabolic assessment as soon as possible (suggested to collect two consecutive 24-hour urine samples and perform serum metabolic tests), and on this basis, strengthen targeted prevention strategies; for those predicted to be at low risk, the routine metabolic assessment and follow-up management should still be followed as per the guidelines, but stratification can be made in terms of resource allocation and educational intensity. It is important to note that this model does not replace comprehensive metabolic assessment but is only used to assist in optimizing the decision-making process and resource priority of “early recurrence assessment-individualized prevention”.It is worth noting that in this study, hyperoxaluria showed a statistically protective effect on “stone composition change” in the multivariate model. This result should not be interpreted as hyperoxaluria having a protective effect on stone formation or recurrence; it may reflect that the continuous oxalate load makes patients more likely to continue along the calcium oxalate-dominated stone formation pathway, thereby reducing the probability of “major component transformation”. On the other hand, this phenomenon may also be related to the limited sample size, incomplete metabolic assessment, measurement errors, and residual confounding, suggesting that the model’s estimates for some variables may be unstable. Therefore, the biological interpretation of this association needs to be further confirmed in larger samples, multi-center external validation, and a more complete metabolic data framework. Conclusion In this cohort of patients with recurrent urolithiasis, shifts in stone composition were observed as a multifactorially driven, dynamic process-specifically associated with hypocitraturia and disturbances in magnesium metabolism. The random forest model developed herein achieved robust discrimination and predictive accuracy under rigorous internal validation (including bootstrap optimism correction). SHAP-based interpretability analysis identified hyperoxaluria, patient age, and baseline stone composition as the top three contributors to prediction. Clinically, the model serves as a decision-support aid for risk-stratified management following recurrence in outpatient settings, informing the prioritization and personalization of preventive interventions. Nevertheless, its generalizability and real-world clinical utility require prospective validation in larger, multi-center external cohorts-and, where feasible, temporal validation via time-split designs. Critically, stone compositional analysis remains a complementary diagnostic modality; it does not supplant comprehensive metabolic evaluation, which constitutes the cornerstone of evidence-based, long-term stone recurrence prevention. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 (19.3KB, docx) Acknowledgements We would like to thank all participants in this study. Abbreviations AUC Area under the receiver operating characteristic curve BMI Body mass index CTU Computed tomography urography DCA Decision curve analysis EAU European Association of Urology HbA1c Glycated hemoglobin HPLC High-performance liquid chromatography ISE Ion-selective electrode CLIA Chemiluminescent immunoassay TC Total cholesterol TG Triglycerides LDL-C Low-density lipoprotein cholesterol HDL-C High-density lipoprotein cholesterol CFU/mL Colony-forming units per milliliter ESR Erythrocyte sedimentation rate PTH Parathyroid hormone UTI Urinary tract infection OR Odds ratio CI Confidence interval MSE Mean square error SVM Support vector machine GBM Gradient boosting machine XGBoost Extreme gradient boosting KNN K nearest neighbor Adaboost Adaptive boosting CatBoost Classification improvement SHAP SHapley Additive exPlanations LightGBM Light gradient boosting machine Author contributions The experiment was designed by Mierzhati Maimaiti, Waresi Xianmuxiding and Guojie Yu. Mierzhati Maimaiti, Waresi Xianmuxiding, Xinjian Chen, Mahemutijiang Atawula, and Guojie Yu carried out the study. Mierzhati Maimaiti and Waresi Xianmuxiding collected the data and did the statistical analysis. Mierzhati Maimaiti, Waresi Xianmuxiding, and Guojie Yu wrote the manuscript, and all authors helped with reviewing and editing it for intellectual input. Funding This research was supported by the 2025 Autonomous Region Medical Service and Security Capacity Improvement (Medical and Health Institution Capacity Building) Program (Document No.: XWYYH [2025] No. 14). Data availability The datasets generated and/or analyzed during the current study are not publicly available due to participant privacy and confidentiality requirements but can be made available from the corresponding author upon reasonable request. Declarations Competing interests The authors declare no competing interests. Ethics approval and consent to participate This study was conducted in accordance with the ethical principles outlined in the Declaration of Helsinki and received formal approval from the Institutional Review Board (IRB) / Ethics Committee of the Friendship Hospital of Urumqi (Approval Number: 2025N0.46).​. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Mierzhati Maimaiti and Waresi Xianmuxiding contributed equally to this work. References 1. Calculi Group of Chinese Urology Association. Chinese Urolithiasis Alliance. Chinese expert consensus on metabolic evaluation and recurrence prevention of urinary calculi. Chin. J. Urol. 41 (12), 881–886 (2020). [ Google Scholar ] 2. Na, Y. Q. et al. Chinese Guidelines for Diagnosis and Treatment of Urological Diseases (2022 Edition) . Urinary Calculi Volume (People’s Medical Publishing House, 2022). 3. Zhang, L. et al. Epidemiological characteristics of urolithiasis in China: A cross-sectional study. BMC Urol . 21 (1), 324​ (2021). 4. Scales, C. D. Jr, Smith, A. C., Hanley, J. M. & Saigal, C. S. Urologic Diseases in America Project. Prevalence of kidney stones in the United States. Eur. Urol. 62 (1), 160–165 (2012). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Wang, J., Li, J. X. & Zhang, X. Research progress on risk factors and prevention strategies of urinary calculi recurrence. Chin. J. Urol. 44 (5), 395–398 (2023). [ Google Scholar ] 6. Evan, A. P. & Worcester, E. M. Pathophysiology of nephrolithiasis. Compr. Physiol. 7 (3), 1003–1018 (2017). [ Google Scholar ] 7. Liu, T. Z., Chen, Z. Q. & Ye, Z. Q. Analysis of urinary stone composition and its clinical significance. J. Clin. Urol. 34 (8), 667–670 (2019). [ Google Scholar ] 8. Krambeck, A. E. et al. Changes in kidney stone composition over time. Kidney Int. 81 (10), 1020–1025 (2012). [ Google Scholar ] 9. Cheng, Y. & Chen, L. J. Research progress on the correlation between metabolic disorders and recurrence of urinary calculi. Natl. J. Androl. 28 (7), 655–659 (2022). [ Google Scholar ] 10. Wang, H. et al. A logistic regression model for predicting kidney stone recurrence. Urol. J. 17 (4), 365–370 (2020). [ Google Scholar ] 11. Lee, J. S. et al. Machine learning-based prediction of kidney stone composition using clinical parameters. Sci. Rep . 11 (1), 18962​ (2021). 12. Chen, G. F., Huang, J. & Xu, K. W. Application of support vector machine model in predicting recurrence risk of urinary calculi. Chin. J. Urol. 41 (3), 192–196 (2020). [ Google Scholar ] 13. Liu, J. H. & Zhou, L. Q. Challenges and prospects of precise prevention and treatment of urinary calculi. Natl. Med. J. China . 103 (12), 889–891 (2023). [ Google Scholar ] 14. Vickers, A. J. & Elkin, E. B. Decision curve analysis: A novel method for evaluating prediction models. Med. Decis. Mak. 26 (6), 565–574 (2006). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 15. Collins, G. S., Reitsma, J. B., Altman, D. G. & Moons, K. G. M. Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD): The TRIPOD statement. J. Clin. Epidemiol. 68 (2), 112–121 (2015). [ DOI ] [ PubMed ] [ Google Scholar ] 16. Xu, C. B. et al. From stone removal to etiology-oriented prevention: Paradigm shift in urolithiasis management. Chin. Med. J. 137 (8), 956–963 (2024). [ Google Scholar ] 17. Baowaidan, F. et al. Incidence and risk factors for urolithiasis recurrence after endourological management of kidney stones: A retrospective single-centre study. Prog. Urol. 32 (8–9), 601–607 (2022). [ DOI ] [ PubMed ] [ Google Scholar ] 18. Skolarikos, A. et al. Metabolic evaluation and recurrence prevention for urinary stone patients: An EAU guidelines update. Eur. Urol. 86 (4), 343–363 (2024). [ DOI ] [ PubMed ] [ Google Scholar ] 19. Liu, Y. et al. Risk factors for urinary stone formation: A population-based study. BMC Nephrol. 21 (1), 389 (2020).32894077 [ Google Scholar ] 20. Sultan, M. I., Hammad, M. A. M. & Youssef, R. F. Sex and age variation for nephrolithiasis risk factors on 24-hour urine metabolic evaluation: A real-world single-center retrospective analysis. Urology 204 , 34–40 (2025). [ DOI ] [ PubMed ] [ Google Scholar ] 21. Ferraro, P. M., Taylor, E. N. & Curhan, G. C. 24-Hour urinary chemistries and kidney stone risk. Am. J. Kidney Dis. 84 (2), 164–169 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Skolarikos, A. et al. European Association of Urology Guidelines on the diagnosis and treatment of urolithiasis. Eur. Urol. 88 (1), 64–75 (2025). [ DOI ] [ PubMed ] [ Google Scholar ] 23. Pearle, M. S. et al. Medical management of kidney stones: AUA guideline. J. Urol. 192 (2), 316–324 (2014). [ DOI ] [ PubMed ] [ Google Scholar ] 24. Li, M. et al. Changes in stone composition during recurrence: A retrospective cohort study. Urol. Res. 49 (5), 673–680 (2021). [ Google Scholar ] 25. Chen, B. et al. Dynamic changes of urinary metabolites in recurrent stone patients. J. Chromatogr. B . 1201 , 122890 (2022). [ Google Scholar ] 26. Peterson, C. A., Adams-Huet, B. & Assimos, D. G. Stone composition variability in recurrent nephrolithiasis. Urology 143 , 231–236 (2020).32439556 [ Google Scholar ] 27. Morgan, T. M., Pearle, M. S. & Goldfarb, D. S. Urinary pH and stone composition: Clinical implications. Curr. Opin. Nephrol. Hypertens. 30 (4), 321–327 (2021). [ Google Scholar ] 28. Johnson, E. M., Reynolds, T. M. & Fathallah-Shaykh, H. M. Urinary tract infection and stone composition transformation. Infect. Dis. Obstet. Gynecol . 2022 , 5489761 (2022). 29. Pak, C. Y. Citrate and nephrolithiasis: Prevention and treatment. J. Am. Soc. Nephrol. 31 (9), 2001–2009 (2020). [ Google Scholar ] 30. Hoppe, B., Kuhlmann, M. K. & Wingen, A. M. Hypocitraturia: Diagnosis and management. Nat. Rev. Nephrol. 17 (8), 489–502 (2021). [ Google Scholar ] 31. Dominguez-Gutierrez, P., Rodriguez, M. & Canales, B. K. Role of magnesium in nephrolithiasis. Minerals 12 (3), 345 (2022). [ Google Scholar ] 32. Worcester, E. M. & Coe, F. L. Calcium oxalate nephrolithiasis and hyperoxaluria. Nat. Rev. Nephrol. 16 (11), 661–674 (2020). [ Google Scholar ] 33. Traxer, O., Huet, B. & Poindexter, J. Hyperlipidemia and kidney stones: A systematic review. Urology 154 , 20–27 (2021). [ Google Scholar ] 34. Torricelli, F. C. M. et al. Dyslipidemia and kidney stone risk. J. Urol. 191 (3), 667–672 (2014). [ DOI ] [ PubMed ] [ Google Scholar ] 35. Inci, S., Akgul, E. & Turkcapar, N. Metabolic syndrome components and urinary stone composition. Urol. J. 18 (3), 298–303 (2021). [ Google Scholar ] 36. Breiman, L. Random forests. Mach. Learn. 45 (1), 5–32 (2001). [ Google Scholar ] 37. Moe, O. W. Kidney stones: Pathophysiology and medical management. Lancet 367 (9507), 333–344 (2006). [ DOI ] [ PubMed ] [ Google Scholar ] 38. Shastri, S., Patel, J., Sambandam, K. K. & Lederer, E. D. Kidney stone pathophysiology, evaluation and management: Core curriculum 2023. Am. J. Kidney Dis. 82 (5), 617–634 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 39. Ishigami, K., Yamashita, Y. & Okada, Y. Machine learning for predicting kidney stone recurrence. J. Urol. 206 (3), 523–529 (2021). [ Google Scholar ] 40. Zeng, G. et al. International Alliance of Urolithiasis (IAU) guidelines on the metabolic evaluation and medical management of urolithiasis. Urolithiasis 51 , 4 (2023). (article number 4). [ DOI ] [ PubMed ] [ Google Scholar ] 41. Castle, S. M., Cooperberg, M. R., Sadetsky, N., Eisner, B. H. & Stoller, M. L. Adequacy of a single 24-hour urine collection for metabolic evaluation of recurrent nephrolithiasis. J. Urol. 184 (2), 579–583 (2010). [ DOI ] [ PubMed ] [ Google Scholar ] 42. Altunhan, A. et al. Artificial intelligence in urolithiasis: A systematic review of utilization and effectiveness. World J. Urol. 42 (1), 579 (2024). [ DOI ] [ PubMed ] [ Google Scholar ] 43. Gong, J. B. et al. Keto-enol tautomers as intrinsic crystal growth inhibitors for urate stones. Nat. Commun. 14 (1), 1245 (2023).36871075 [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Supplementary Material 1 (19.3KB, docx) Data Availability Statement The datasets generated and/or analyzed during the current study are not publicly available due to participant privacy and confidentiality requirements but can be made available from the corresponding author upon reasonable request. Articles from Scientific Reports are provided here courtesy of Nature Publishing Group ACTIONS View on publisher site PDF (2.4 MB) Cite Collections Permalink RESOURCES Similar articles Cited by other articles Links to NCBI Databases Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top

Related documents

• Filtration-Based Learning of Multiscale Shared Structures for Multiple Functional Predictors#163152
• Multiple instance fine-mapping: Predicting causal regulatory variants with a deep sequence model#323627
• Beyond the reducing valve: towards a computational neurophenomenology of altered states via deep neural networks.#258341
• Beyond the Reducing Valve: Towards a Computational Neurophenomenology of Altered States via Deep Neural Networks#29849
• Neural Bradley-Terry Rating: Quantifying Properties from Comparisons#478178